November 9, 2026
Same Story, Different Sides
How AI’s Judgment Changes Depending on Who’s Asking
We gave AI assistants the same fictional disagreement from opposite perspectives, then asked them to take our side. Some replies grew more certain about blame, inferred motives the texts did not establish, or changed the order of events. Other replies kept important limits. This early case study documents what changed, what resisted pressure, and what we want to test next.